KVM on AMD: verify remediation where the virtual machines are hosted
The KVM flaw calls for identifying who operates the host and nested virtualization. A business response depends on verifying the configuration and the remediation actually in use.
Czytaj po portugalskuIn this article
- Start with the confirmed technical condition
- Separate research from evidence about your environment
- Create a record for each group of hosts
- Connect the fix to the package and running kernel
- Give temporary measures a deadline and an exit condition
- Frequently asked questions
- Related reading
- Sources and references
By Pentesty
Start with the confirmed technical condition
The November 20, 2026 Register report prompted this analysis. The Linux CNA record, published on November 18 and updated on November 19, rates CVE-2026-99312 as high: CVSS 3.1 8.8. The condition is KVM on AMD with nested virtualization.
The kernel change fixes memory context reset on the SVM path and distinguishes VMX behavior. That calls for a review with a known architecture and configuration. An inventory containing only Linux or virtual machine is insufficient to decide which hosts require action.
Separate research from evidence about your environment
In Terra Security’s research, Oren Yomtov reports reading host memory from a DigitalOcean VM, with the write test performed on his own server. The result is experimental; this analysis has no provider confirmation of an incident.
For the business, the practical question is who can confirm each layer’s condition. On infrastructure it operates, the operations team can connect hardware, the virtualization module and the running kernel. For a contracted service, identify the environment and ask the platform owner for a specific response about applicability and maintenance status. A generic statement that systems are updated does not establish that connection.
Create a record for each group of hosts
Group servers by the service they support and their operational owner. Record architecture, distribution, kernel package, running version and nested virtualization use. Identify who can change the configuration and which workloads depend on that feature. The result should distinguish a verified host, one lacking sufficient evidence and an environment outside the condition being assessed.
Do not attempt to reproduce an escape on shared infrastructure merely to complete the record. Assessment requires an authorized scope and methods consistent with the contract. The earlier KVM on ARM64 analysis helps organize guest and host responsibilities, but concerns a different architecture and CVE. Its technical conclusions should not be transferred automatically to this case.
Connect the fix to the package and running kernel
The CNA record lists fixes in 6.12.118, 6.18.64, 7.3.5 and 7.4-rc2. It also lists affected entries from 5.15.81 before 5.16, and from 6.0.11 before 6.1. Check the history and backports of the distribution’s supplied package.
Plan maintenance with the people accountable for availability. Acceptance should connect the applicable advisory to the approved package and confirm the kernel loaded after the change. Record workload checks, timing, the owner and remaining exceptions. The Linux kernel qualification and reboot guide develops this evidence practice further.
Give temporary measures a deadline and an exit condition
Terra identifies disabling nested virtualization in the kvm_amd module as an alternative when updating is not possible. Before that change, identify dependencies, approve the maintenance window and verify the effective configuration. An intention recorded in a ticket is not evidence that the protection is already deployed.
Keep a short exception list: service, host or group, adopted measure, missing evidence, owner and next review. Separate technical acceptance from the business decision about temporary continuity. A useful conclusion explains which environments were checked, how remediation was established and what still requires action.
When a service involves several suppliers, keep the same record for each contractual relationship. A response from the operating system distributor does not replace confirmation from the host operator. Route discrepancies to the technical owner before closing the task, preserving the decision history.
Frequently asked questions
Does updating Linux inside the VM fix the host?
It does not establish host remediation. The team responsible for virtualization must verify its own package and running kernel.
Is every virtual machine on AMD exposed?
No. Verify the documented technical requirements before classifying the environment.
Does the research prove a DigitalOcean incident?
No. Distinguish a controlled demonstration from provider confirmation of an incident.
What evidence closes the maintenance task?
A verifiable connection between the host, applicable fix, approved package, active kernel and workload checks, with an accountable owner and documented exceptions.
Sources and references
- The Register: November 20, 2026 report
- Linux CNA: CVE-2026-99312 record updated November 19
- Terra Security: Oren Yomtov research published November 18
- Linux kernel: fixing commit in the stable maintainer’s mirror
Original Pentesty analysis, published November 20, 2026. The proposed scenarios are editorial guidance, with scope and validation to be defined for each environment.